← Back to Downloads

EU G-SIB Stress-Test Dataset — v1 · April 2026

An open, reproducible practitioner dataset spanning the 2018, 2021, 2023 and 2025 EBA EU-wide stress-test cycles, covering 18 European G-SIBs on a harmonised template. Peer benchmarks on starting CET1, CET1 depletion, risk parameters (PD, LGD, coverage), credit losses, NII under stress, operational-risk charges, and the corresponding ESRB adverse macro scenarios. Built for analysts who need to do the arithmetic themselves.

Version
v1.0
Released
April 2026
Banks
18 G-SIBs
Cycles
2018 · 2021 · 2023 · 2025
Download the dataset (.xlsx)

What’s inside

Nine tabs, each a normalised long-format table keyed on bank_id · cycle · scenario · year. No pivot tables, no merged cells, no formatting-as-data. If you can load a CSV, you can load this.

01

Scenarios

ESRB adverse & baseline macro paths per cycle — GDP, unemployment, HICP, residential & CRE prices, equity, sovereign spreads, 3m/10y yields, FX. 28 EEA jurisdictions plus US, UK, CN, JP.

02

Starting CET1

T-0 CET1 ratio (transitional & fully loaded), CET1 capital, RWA by risk type, leverage ratio — as declared in the EBA transparency templates for each cycle.

03

CET1 Depletion

End-point CET1 under adverse, depletion in bps from T-0, minimum CET1 across the 3-year horizon, distance to MDA trigger. 2025-cycle figures adjusted for CRR3 Output Floor phase-in.

04

Risk Parameters

PD, LGD, coverage ratios by portfolio (corporate, SME, retail mortgage, retail other, sovereign), T-0 and projected under adverse. Stage 1/2/3 migration where disclosed.

05

Credit Losses

3-year cumulative credit losses, annual impairment, cost-of-risk in bps on average gross exposure. Broken out by IFRS 9 stage and portfolio for the cycles where EBA disclosed that granularity.

06

NII under Stress

NII path under adverse vs baseline, re-pricing effect, behavioural-deposit assumptions where disclosed, IRRBB-adjacent shocks. Cycles from 2021 onward use the constrained-balance-sheet methodology.

07

Op Risk

3-year cumulative operational-risk losses — headline, conduct-risk overlay (where split out), and the EBA floor in basis points of operational RWA.

08

Peer Benchmarks

Pre-computed medians, p25 / p75 and country-cohort bands for every key variable. Lets a single bank position itself against EU G-SIB peers in one lookup.

09

Methodology

Defaulting rules, sources per cell, version log, known gaps & caveats. Every number either traces back to a primary EBA/ESRB/Pillar 3 source or is labelled as an Ezelman practitioner estimate.

Methodology & data quality

We take data quality seriously because a peer-benchmarking dataset is only as good as its worst cell. Three control loops are in place.

A · Primary sources

Sourced from the public record

Every figure traces to a named, dated primary document:

  • EBA EU-wide stress-test transparency templates — 2018, 2021, 2023, 2025 cycles
  • ESRB adverse-scenario annexes — macro paths, FX shocks, yield curves
  • Individual bank Pillar 3 disclosures for gaps the EBA templates don’t cover
  • ECB SREP public Pillar 2 capital-requirement ranges, where publicly disclosed
B · Data-quality controls

Three independent checks per row

Before a row is admitted to the master tables, it passes:

  • Row-level cross-check — every cell reconciled against two independent sources (e.g. EBA template vs Pillar 3) where both exist
  • Sample-completeness audit — every bank-cycle pair has a completeness score; scores < 90% flagged visibly in the file, not quietly imputed
  • Source tag on every cell — a parallel reference sheet carries the specific EBA table / Pillar 3 page number per cell; where we estimate, the cell is tagged EST-EZ
C · Defaulting rules

Completion without silent imputation

Where an EBA template cell is blank for a specific bank, we apply a documented and reproducible fallback:

  • Country-cohort median regression for missing banks in a cycle — predicts from peer-cohort risk profile with the residual reported
  • Sequential fallback for blank cells — prior-cycle value → Pillar 3 disclosure → country-median regression → Ezelman estimate
  • No silent imputation — every filled cell carries its fallback tier; a user can re-filter to a strictly sourced subset at any time

Why we ship this

The Big-4 build peer-benchmarking datasets worth seven figures in chargeable hours and never release them. The big strategy firms release framework decks but not data. The result is that every risk function in Europe pays someone, twice a year, to scrape the same EBA templates into the same shape — and nobody shares back. Ezelman is on the other side of that position. We ship what we build. This dataset is part of a wider commitment we call on the record: the work we do for clients we can package, the packaged work we can share, and the positions we take we sign — dated, and corrected in public.

Update cadence & licence

Updated on every EBA cycle (next material refresh: Q3 2027, tracking the 2027 cycle results disclosure). Interim patches released whenever a reviewer flags an error we can verify.

Licence
CC BY 4.0 — use freely with attribution to Ezelman.
Update cadence
One major release per EBA cycle + point patches on verified corrections.
Corrections
research@ezelman.com — every corrected cell is versioned in the file log.

Use it, challenge it, email us corrections.

The dataset is only as good as the community that stress-tests it. If you spot an inconsistency, a missing Pillar 3 figure, or a cohort-median regression you can improve on, write to us. Corrections are published in the changelog under the submitter’s name if they want it there.

Download v1 (.xlsx) → Email research@ezelman.com
Follow Ezelman on LinkedIn — Regulatory insights read by 2,000+ risk professionals across Europe & the GCC Follow →